ETH Zurich test revives the 'too hot' climate models and moves policy warming to 3.25C

A group at ETH Zurich has proposed a different way of deciding which climate models to trust, and it reverses the previous verdict: models that were set aside for simulating more warming than was observed are the ones that best match satellite measurements of energy entering and leaving the atmosphere ^[1][11]. For a given emissions path, the team expects warming to be 25 per cent higher than current projections, according to lead researcher Gergana Gyuleva ^[2].

The number that matters to anyone building plans is the translation. The Climate Action Tracker currently projects 2.6C of global surface warming by 2100 if countries implement their existing policies ^[3]; on the ETH Zurich results, the same policies deliver 3.25C ^[4]. That is 0.65C of additional warming attached to no new emissions assumption at all, only a change in how model skill is judged ^[1].

The disagreement being adjudicated is old and wide. Some models put the short-term surface warming from a doubling of CO2 at around 1.6C, others at as much as 3C ^[6]. "The range is big, and it really matters," Gyuleva told New Scientist ^[7]. The standard filter has been hindcasting: feed models a century of emissions and compare their output with what happened ^[8]. For the last IPCC report, many models simulated more warming than had occurred and were discounted, leaving a range of 1.2C to 2.4C for a doubling of CO2, best estimate 1.8C ^[9][5].

Gyuleva's team attacks that filter on two fronts. First, they strip out natural variability such as La Nina, which brings cooler deep water to the surface ^[10]. Their results indicate variability suppressed temperature rises between 1981 and 2014, the window that includes the early-2000s warming hiatus and the window used in the earlier papers ^[12]. "It was really bad luck," Gyuleva says ^[13]. Correcting that bias alone raises the estimated sensitivity ^[14].

Second, they grade models on radiation rather than temperature. Satellite instruments measure incoming shortwave and outgoing longwave radiation, and the gap between the two is the most fundamental measure of warming ^[15]. Those records only start in 2001, which is why this test is newly available ^[16]. Models that best match the observed radiation trends project much more warming ^[11]. Combining both assessments gives 1.9C to 2.6C for a doubling of CO2, best estimate 2.25C ^[17]. The new lower bound sits 0.1C above the IPCC's old central estimate ^[2], and the central estimates differ by exactly the claimed 25 per cent ^[3].

The outside reaction is qualified rather than dismissive. Trude Storelvmo at the University of Oslo calls it a sound approach ^[18]. Drew Shindell at Duke calls the removal of natural variability a valuable step forward but is not convinced the shortwave and longwave trends are a good way to assess models ^[19]. Steven Sherwood at the University of New South Wales calls the conclusion plausible, while noting the work still rests on climate models and none of them reproduces the key phenomena well ^[20][21]. Gyuleva herself says the results need confirmation from other sources ^[22].

Watch whether the radiation-trend test survives Shindell's objection, since it is the load-bearing half of the argument. Watch also Sherwood's point that recent steep temperature rises will push estimated ranges up regardless of method ^[23]; if he is right, 2.6C-based design assumptions are exposed either way.